Towards pulmonary vessel separation

Abstract

Pulmonary artery-vein segmentation and separation plays an important role by guiding doctors’ surgical planning. However, automation of this process has not yet been fully achieved. There are plenty of studies that achieve fairly accurate results. However, most of the available algorithms rely on highquality CT scans, which are not always available in everyday practice. High quality for CT means (1) contrast-enhanced CT scans with (2) well-timed contrast material injection, (3) thin slice thickness, and (4) the application of filtering during the reconstruction to reduce the amount of streaking artifacts. In this work, we present a set of algorithms to improve an existing system for artery-vein separation, with the aim of making it more robust on CT scans of typical quality. Furthermore, their results are investigated, discussing their strengths and current difficulties.